Session pool management in Machine Learning Server 9.3.0 RRS feed

  • General discussion

  • I am using session pools (generic and Dedicated) in my application with deployed scripts (R & Python) in a windows environment.  Everything works beautifully as far as execution (normal, batch and asynchronous).  But when I started monitoring the pools to see how many processes were out there (this is windows :-)), and when I did, I saw the processes go up just as I expected, but when the code execution was complete the python sessions went back down to the minimum, the R sessions did not.  They don't appear to ever recycle. Is there a setting somewhere to recycle them faster when they're not used, like the python sessions?

    My settings for the R and Python session pool are initial: 10, max: 25.  This is just my local development laptop.

    I ran a batch that executed 13 different calculations in batch with 3 datasets for each.  The session count went up to about 18 sessions for python and R. 

    Once the program was finished however, the python went back to 10:

    ** Image to be provided when the forum allows....

    When I look at the R sessions, they're still at the expanded count:

    ** Image to be provided when the forum allows....

    This is over an hour since the last execution against the server.  It should recycle and release these processes, correct?  If not you would just end up with a HUGE generic session and you could eventually exhaust server resources, if there are users or code that are creating dedicated session pools or individual sessions to interact with ML Server.  Not sure if anyone else has run into this, but we have to load test and provide proof of our application footprint before it can go to implementation.

    Any suggestions, links or hints someone may have for this would be greatly appreciated.



    Friday, October 5, 2018 5:21 PM